EP273 From CISA to Cloud: AI Assurance, Concentration Risk, and the New Regulatory Frontier

Security. Hi there. Welcome to the Cloud Security Podcast by Google. Thanks for joining us today. Your host here is ever myself, Tim Peacock, PM in Google SecOps, and of course, Anton Chuvakin, member of senior staff in Google Cloud's Office of the CISO. You can find and subscribe to the show wherever you get your podcasts, as well as at our website, cloud.withgoogle.com/cloudsecurity. If you enjoy our content and want it delivered to you piping hot, please do hit that subscribe button. You can follow the show, argue with us and the rest of our Cloud Security Podcast listeners on the Google Cloud Security Community page. Anton, we have a guest today who did the best job of kind of like Jedi mind tricking us. We tried to Jedi mind trick her with a left or right, and she did the best job, I think, since Kathryn Hsu, who came on the show like 2 years ago now, of saying it's neither left nor right. It's actually behind you, morons.

Yes. And this was like a super impressive move. But also, the topic is actually way cooler than you'd get from reading the headline when I would invent the headline for the episode. It's compliance.

Maybe you should invent a better headline, though.

Oh, yeah. That's a good point. At this point, it's not too late to make a better headline.

It's not too late. No, this is a super fun episode, though. And I think we had no lawyer cats fall out of the ceiling, which we're very fortunate about as well.

Yeah, no lawyer cats. But there was a brief moment when I felt like they're meowing in there.

Ah, you heard them getting closer on little cat's paws.

Yeah, yes, exactly. A little lawyer cat claws.

I have heard lawyers have many clauses.

Oh no, a pun.

All right, listeners, with that, let's turn things over to today's guest. Today we are joined by Jeanette Manfra, VP, Head of Risk and Compliance for Google Cloud. Jeanette, thank you so much for joining us today. You and most of our friends in Office of the CISO talk about cloud computing as a way to simplify security. Which, as somebody who spent many years building those features, thank you. As a CISO dealing with global regulators, though, how does cloud make that security versus privacy conflict more or less intense? Because now you've outsourced your security to some extent. So, how does that change the picture for a CISO?

Yeah, great question. And thanks for having me. And thanks for your podcast. I am a regular listener. Oh, thanks.

Thank you.

So, So, I actually, I talk a lot about my own experience moving into the cloud and coming into Google. And so, to kind of provide some of that context there, I spent my whole career before Google in the US government, which is well known for its advanced technology. And no, just kidding.

Wait a minute, that was a joke.

You almost had us there, by the way. You almost got us. I was this close to nodding and said, yep, yep, yep, they are. Think about DARPA.

Very well known for advanced technology there. Yeah, sure, sure, sure, go on.

It's defining elements and low cost and, you know, all the great things.

And on-time delivery of projects, right?

Yes, absolutely.

Yes, yes.

And I spent my whole time in the security field, variety of different roles there. And as we started to look at cloud, of course, as security professionals and in the government, especially in national security space, it was like, oh wait, this is terrifying, this is new, this is different, we don't like this. And then we threw out a couple of administrations over the past, I don't know, 15 years or so, started looking at, well, actually cloud provides a tremendous amount of efficiency benefits, cost savings for, you know, taxpayers, all those wonderful things. And so started to get more drawn in as a security professional and started to learn a lot more about it. And one of the first things that I realized is how much security capability you get Now, that doesn't mean it's, you know, just you can kind of close your eyes and say, oh, everything's better when I move to the cloud. You have to be like smart about it. But I started to realize the benefits and, you know, what some people might— I hear people talk about like the democratization of security. And it's because I spent so much time when I was at CISA, the Cyber and Infrastructure Security Agency, a lot of the work we were doing was with organizations who couldn't afford the best security. And they were highly vulnerable. They were very critical to the functioning of government, to the functioning of society. And so, I really actually started to become a believer in the power of cloud done right, done with transparency, done with audibility, all those good things, but in recognizing that you get those, you know, those economies of scale for security, in addition to all the rest of the great product features that you get. And that was actually a key reason why I came to Google Cloud, because I felt that there was so much power to bring security to the world in a way that historical technologies weren't able to do that.

I love that. I love the, like, translation of government mission orientation to Google mission orientation of bringing security technology to the world that they couldn't have before. Because that's That's really true, isn't it?

Yeah, and it is. And one of the, a lot of what I've done here, of course, now in my role in risk and privacy and compliance, I spend a lot of time thinking for ourselves internally, of course, but also talking to customers and whether they're CISOs, chief risk officers, chief compliance officers. In your question, you talked about how does cloud help or challenge the constraints that many regulated organizations are under, and especially when they're multinational corporations, that have to deal with a whole patchwork of different regulations. Even organizations that aren't multinational, oftentimes they have to deal with a patchwork just within their own country. And looking at you, US state laws. But the—

Ah, federalism, isn't it great?

Yeah, so if you take the benefits that cloud brings from a security perspective, from, again, even from a privacy perspective, from a reliability perspective, all of those things is critical. And now at the point where Most organizations, especially in these regulated, more risk-averse, we'll call them, sectors, they want these things. But what's always important is to make sure that you're bringing the regulators, your internal risk people, your internal compliance people, your internal security people along so that they can, you know, you're not just like deploying a whole thing and then they're sort of left struggling to figure out, well, how does this apply? And how do I sort of make sure that I'm still compliant? How do I make sure I do the right things? In the end, regulators, Regulators, compliance, risk, security folks, they're usually aligned on the outcomes, right?

Yep.

There's always that outcome. I've never talked to a regulator that didn't have the same sort of security goals that I do. How you get there is different when you're talking about cloud, when now, when you're talking about AI. Sometimes how you get to those security outcomes or those sovereignty outcomes or those privacy outcomes can be different. And you need to be able to have that open conversation. And importantly, you need to be able to enable a regulator or a customer to verify what you're saying through technological means and all of that. And that's so that transparency is important. But I don't think that cloud as a sort of a, I guess I would say, meta-grouping of technologies or AI in particular changes the paradigm per se.

Hmm.

If you're focused on what the outcomes are. Oftentimes, of course, compliance folks in particular get focused on the, what you're doing to get there.

Yeah.

And so, you have to sometimes reframe the conversation. But no, I don't think it necessarily makes the security versus privacy or some of these other things that people talk about a more intense conversation. It just makes it a different conversation. And it is also different in terms of what our responsibility is and what our customers are. I'll pause there.

So this is a very elegant answer because it, instead of, despite our attempt to sort of maybe not trick you, but kind of say, is it left or right? And Jeanette is like, grabs the steering wheel and like, actually, it's the other direction altogether. So, I really like this, but I want to push very gently in one direction. What about the centralization argument? Cloud does centralize. Well, I mean, cloud also decentralizes, so I may defeat my own argument in a moment, but cloud does centralize, does aggregate, and some people say that it does invite scrutiny because it's in some sense more centralized. So, does it make it a bigger target for regulatory scrutiny, or is there some other dimension to this?

Yeah. Well, I think there's kind of multiple layers to that centralization. There's the, like, the sector itself. Is it, you know, too concentrated in just a few providers? That is a conversation that is had often. Are you, as a user of cloud, too sort of dependent on one provider? And then there, you know, goes even further, like, are you too dependent on one region within that cloud? Like, there's a lot of ways to kind of think through that centralization. And I would call it less decentralization and more of a concentration problem.

Mm-hmm.

And there's a risk in anything, right? You want to make sure that things are diversified and that you understand when you're thinking about reliability and resilience in particular, whether that's cybersecurity resilience or, you know, other types, that you understand, you know, what is most critical to your organization and how are you building resilience into that. You know, I don't wanna presuppose what any customer, how they might think. We do spend a lot of time working with customers to help them think through that. Increasingly, regulators are very concerned about this. This is, you know, you see the Digital Operational Resilience Act in the EU. You have in the US the banking industry increasingly looking at this. And to me, the term that I really like to hear is what the, in, in DORA is around this operational resilience, which is something that is known in the banking sector in particular for a long time, but I would say in the tech sector, especially infrastructure is also well known, is the idea of, you know, you have these kind of system resilience, but then you also need to look bigger. How does the financial system, to pick on them, as a whole, how do we ensure systemic resilience? And then how do you do that through understanding your third party, fourth party, et cetera? Again, not unique to the financial sector. It is just one where there's a little bit more regulatory activity focused in that space. And so, I do think that there are always multiple levels of concentration risk that everyone should be thinking about, whether that's as a society, whether that's as industries, whether that's as an individual in how you're managing what's most important to you. And that just really means building in an awareness of where you have those potential single points of failure. We do that all the time in Google. How do we manage those single points of failure? How do we make sure our customers are aware of that, and how do we help them manage? And then how do you think about these broader critical systems, not necessarily technology systems, but broader systems, and how do we build resilience in overall as a, what I maybe call ecosystem, if you will. So, I think that's where you hear people talk most from a regulatory space when they talk about centralization or concentration.

There's so much to dive into with that. I don't even know where to go. I think I just have to go to the next question because I can't process that fast enough to give a good follow-up.

I just baffled you with my brilliance.

You did. You did. No, no, but it was actually, it actually makes sense. I don't want to, like, summarize it because that's already condensed. I think I really appreciate the answer. And also, to very briefly insert a comment which I didn't have—

Oh, please.

Is that people who are spouting things like concentration risk don't always assume that they know what they're concentrating over. But your point was very crisp. It may be the industry, it may be the community, it may be the society, it may be whatever else. So, it is not as black and white as servers in the basement, low concentration, servers in the cloud, high concentration. Life is more nuanced. Yeah.

I want to tell you a slight story because this is just an area that I've worked on a lot. When I was in the government, one of the things that we looked at was kind of to the point of outcomes versus process. One of the things that we were charged with was understanding and ideally trying to prevent a significant cyber incident from disrupting a major national critical function. And one of the big changes that we did was to look at things from a functions and systems perspective, instead of just, oh, here's the 5 major, you know, organizations or bits of software, whatever it might be. And once you start to decompose and you say, what does it actually take for— again, I'll pick on finance since we're talking about it— what does it take for the financial system to operate? And once you start to decompose those different parts of it, you find interesting single points of failure that you didn't necessarily know about. You also find, and I would say the same thing within your own organization, you also find that there is, you know, common software that everybody is using, and maybe that is a vulnerability that we should be looking at, right? And so, if you look at it from a functions perspective, whether that's in your organization and looking at those workflows and looking at those processes, but even broadening it out, this is where most regulators tend to get concerned about, is those systemic resilience, that interplay between, you know, a cloud provider Other cloud providers, and then the parties that are dependent on it, they're trying to prevent these gaps of a lack of a systemic resilience. And I totally agree with them. And, but, and, but I think we take that role very seriously here and look for ways to enhance the resilience instead of cause additional concentration risk. Yeah.

This week's Economist has a massive article about the systemic concentration risk in the chip industry, actually.

Yeah.

There is a very, very scary article out that's like, yeah, here's trillions of economic risk.

Yeah. No, that is actually an Excellent example. Yeah.

Yeah, chips. But that's actually a much crisper risk than the cloud risk because in that sense, cloud is distributed by nature and there is indeed a small number of plants in a small geographic area. Okay. So, this aside, I want to go to a really fun question. And it's a really fun question because it triggers me on so many levels. I've dealt with this quite a bit and I'm dealing with this now with COSIG, or Conshared Responsibility for AI Security. So, I wanna revisit the shared responsibility model for cloud. And, you know, this question may be a little bit 2012 for some people, but I sense that even with regulators, there's an extra dimension. Like, where does the cloud provider's job end versus where the customer's job begins? We can— I mean, we do address it by plotting different diagrams. They have the blue and the brown. Blue is cloud, brown is the customer, or maybe the opposite. Doesn't matter. The point is that, do regulators buy it? If something blows up, if something breaks, do you think that shared responsibility model is understood by regulators? So, I'm not just trying, hey, Jeanette, explain shared responsibility for cloud security. We sort of got that down. But explain the regulator's view of this. And maybe, if you feel adventurous, add AI to it and see if AI changes it.

Yeah. You know, I think from a— let's just start conceptually. The idea that there is a shared responsibility is important and will endure no matter how the, you know, technology underpinning it might change. The idea that you can both very precisely and consistently state exactly how that manifests in every single deployment, in every single part, like, that is not—

Mm-hmm.

That's not something that you can do at a, like, a generic level.

Yeah.

The one slide that everybody sees the first time they're introduced to cloud, where it's got the PaaS, IaaS, and SaaS, and here's, you know, who does what and all that, that is good for a general understanding. But, you know, in the end, it does get back to what I was talking about earlier, is about risk management, and how every regulator I've spoken to is mostly focused on, is the organization doing proper risk management aligned aligned with industry best practices and all of that. Those are starting to change and modernize in terms of how technology can improve those processes and improve some of the transparency. One of the things that I spend a lot of time trying to get internally as well is there's so much that we can do to make the process of checking your work, you know, the compliance, which is effectively like verifying through whatever means, some third-party audit, some customer audit, some reg— all of these things that are very toilsome. They're point-in-time evidence that somebody is doing the thing that they said they were doing. We can do a lot with technology to improve that real-time awareness of the state of controls or the state of the environment. And I like to think that we're really leading in that space, and there's a lot more we can do. There's a lot more that AI can do as well for both our side and our customers. But if I think fundamentally, it is always a shared responsibility. I like— there's, you know, a term that folks started to use around the shared fate. I do like that a little bit better because I feel like shared responsibility sort of just turned into a bit of like a liability framework, if you will.

Yeah.

And in the end, we're in there with our customers. Yes, do we have to make it super clear what we're going to provide, what our customers are going to provide, and do we often find it useful to bring the regulator into that conversation, whether that's our customers, regulators, or us. Absolutely. If that is an environment that we can have that sort of trusted conversation, it is really, really helpful. And then you get into the details of like, here's how I'm making sure that my infrastructure meets the requirements that you have, and here's how I'm proving it to you. And by the way, regulator, here's, you know, the conversation that we're having, and have that like very clear trusted dialogue between what our customer is doing and what we're doing. I don't think any of that is going to change. I think I think the specifics under which everything devils in the details, right? The specifics of how that manifests does change and is changing with the use of AI and how people are thinking about AI and how people are deploying multi-cloud environments, different, you know, supply chain dependent, all these things, right? They manifest in the precise way that that comes together. But fundamentally, it is a shared responsibility or shared fate if our customer is harmed and you need to be right there with them. As, you know, someone likes to say, both of our names are in the newspapers in the end.

Yeah, right? So, I want to pick on something you said there. That's the perfect segue. The toil was going to take me there anyway, but you mentioned AI for compliance. What is that looking like? Are we really using AI to help translate different regulatory requirements to different audit requirements and then produce evidence? Is that really helping?

We are. Isn't that like a separate episode we should do with Jeanette next?

Oh, wow.

Because it sounds like a really good question for the next half an hour. But let's see if she can answer it. Can we answer it in 5 minutes?

And I have to say, it's very special to get invited for the follow-up episode during the episode. Anton does that like twice a year at most.

I am like very honored. I'm blushing. So I would, yeah, for sure. Actually, this is something I'm really passionate about, so I would love to talk more if we have the chance. Let me just give my attempt to be very brief, which is yes, we are finding it very useful. It is not just like click, run, easy, but we have had a lot of success in using natural language processing capabilities to analyze regulations and to be able to rapidly— as you can imagine, Google is, and our customers, subject to lots and lots of different regulations, and we want to be able to understand what those are to ensure that we have the right linking to controls, evidence, et cetera. We have had a lot of success, but that is largely because I have some very awesome, motivated people who've gone out there and decided I'm going to figure this out. I'm going to learn how to prompt engineering, even some low-code capabilities. And so we are definitely using it, and we are finding success. You still have to have human in the loop. You still have to make sure that the human is checking. But what we have done is to scale it a little bit is once we went through the painful process of figuring out how to best set Gemini up to be able to analyze regulations, we then created standard procedures so that other people on my team can be using it. And so, each time it gets a little bit easier. We have big ideas for other things that we could be doing, but our experience has been go slow, be very scoped in what you're trying to do, and then make sure that you always have, because we're in such a sensitive area, that you always have deep human oversight and checking and validating on what we're doing.

Hmm. I like that quite a bit also. But by the way, I feel like AI will manifest a lot more strongly in the coming Months, years. So, yeah, I almost feel like it has to be an episode, Tim. So, that's—

Jeanette, you're coming back.

Yeah. Okay. So, I think that I need to ask the question that refers to the P word again, the privacy, before we start closing. And this is kind of the—

Check if you're stealing for cats first, Anton.

Yeah, I am checking. And the reason for this joke is that I'm always afraid that lawyer cats would start streaming from my air vent. So, I have this paranoid vision of if I mention the P word privacy on the podcast, that lawyer cats would start meowing. Out of ventilation. So, when we get demands for transparency, we may increase logging, increase certain types of data collection, check in for configurations, check in for other things, deliver more insight, and then people say, no, no, no, no, no, you can't record all that because of privacy. Can you give us at least the ideas for how do we deal with this? Like, I am— this is probably the scariest thing for me because how do we actually combine those requirements?

Privacy and transparency.

Yes.

Logging, auditability, and privacy.

Yeah, that's fair.

And security.

How do you do that? Right. So, I do also run privacy for cloud. And so, in addition to the lawyers, it'll be my own privacy team coming out of the air vents if I—

Oh, boy. Oh, thank you. You really needed to say that, by the way. I am like—

First of all, I think this is— it's a fascinating topic as somebody who actually, I do care a lot about privacy, in addition to meeting those, you know, regulatory outcomes that are different. Privacy is a fascinating thing all on its own. There are lots of different concepts of what different cultures and different countries think about what it means to be private. And that has then consequences in how they regulate privacy and how they do the balance between privacy and security, privacy and transparency. So just sort of setting that up, it is different depending on where you're at and how you deal with it. So there's not like a universal thing that we can say. I would say general guidance when it comes to laws logging, transparency, auditability, all of that. You want to ensure that you have the appropriate privacy filters in place, but you are going to need to adapt that to wherever you operate in whatever region, for sure. We put a lot in place ourselves. We do not train our models on private data. We do not— there's a variety of things that we offer to customers to increase if they have an additional set of restrictions that they'd like to put in place to prevent anybody from accessing it anywhere. There's all sorts of things we can do. There's also one of the things, I have a privacy engineering team who's got all these, you know, really brilliant ideas of thinking about things like differential privacy, thinking about agentic privacy, thinking about how can we achieve these security transparency outcomes by using math. And, you know, there's really interesting that I am not qualified to talk about in depth, but using these differential privacy techniques and others to be able to get the best of both worlds, if you will.

Hmm.

To preserve the privacy whatever those requirements might be, in whatever location you might be, but to be able to still have the transparency, the security, and other outcomes. And so there's work that Google has done already that we've pushed out publicly. There's work that we're continuing to do that we will push out publicly. But I do think that we're starting to get to this place, potentially, where we can have our cake and eat it too, keep people's data private while ensuring that we maintain the benefits of the access to non-private data for security, transparency, and other outcomes. But this is a big grand challenge and one that I'm really excited to be a part of.

That's a very optimistic note to end on. And so as much as I hate to do it, I'm happy this is where we're doing it. I have to ask our traditional closing questions. So one, do you have a piece of advice for other security and privacy leaders to help their teams, not just their companies, but like their teams adopt AI for their practices? And then 2, do you have recommended reading? And it could be anything, just not Anton's blog.

Yeah, I have a lot of recommended reading. Some might not be relevant to this. I'm reading Brothers Karamazov right now.

Oh, nice.

So if you're just sort of in the mood for anything really deep, Russian, dark, but also like—

I get enough of that with Anton.

Yeah, that's good. And then I've been I've been talking a lot. I read this awesome book called Wiring the Winning Organization.

Ooh.

So, it's a mix of, like, learning from tech sector, but also manufacturing practices from the '80s, and all this other stuff. So, how do you get— I really love, like, future of work ideas, which is going to bring me back to your first question. And, you know, how do you empower teams to be more agile, and how do you, as managers and leaders, create the environment for them to be more agile and autonomous, while still working towards common goals, without that sort of rigid decision-making hierarchy process. So—

That's cool.

Throw that out there because I'm a huge fan of that book. And there's a few others in that genre. And then kind of to that, that's a bit how I've been talking to my own team about thinking about risk, privacy, and compliance. And I did, I'll plug my own blog.

Oh yeah, please.

I put a blog out in the Cloud CISO Perspectives, thinking about AI as a strategic advantage for risk and compliance. I acknowledge that a lot of folks are— that makes them nervous because there's a lot of things that we do in this space that are great use cases for agents and other AI capabilities to take over. And the way that I think of it is how do you think of AI as an enabler and accelerator for you so that you are driving the critical thinking, you're driving the orchestration, you're making life easier for, you know, the product teams, the service teams, the mission, whichever industry you're in, and to think of it as an enabler. And if you're not, you know, maybe more tech or AI savvy, which many folks aren't necessarily in this community, there's so much out there to get that training and to not be intimidated by it. Test it out, play around with it, but mostly don't be intimidated by it. Think of it as an enabler for you to get to that kind of next level of risk management. Thanks.

I love that answer. And now I'm picturing you at home playing with, like, an Open Claw on your Mac Mini at home.

Is that what's happening here? I hope not. No comment.

Amazing. Well, listeners, we managed to make it 30 minutes into an episode with a VP of Risk and Compliance at Google, and only at the end did we get a no comment. So I'd consider that a podcasting win for today. Jeanette, thank you so much for joining us. I'm really looking forward to having you back on the show to talk about, I mean, just so much Thank you. And now we are at time. Thank you very much for listening, and of course, for subscribing. Please subscribe so you can get new episodes piping hot. And if you love our content, please drop us a review on your platform of choice.

You can find this podcast on YouTube, Apple Podcasts, Spotify, or wherever you get your podcasts. Also, you can find us on our website, cloud.withgoogle.com/cloudsecurity/podcasts. You can argue with us, You can join us on the Google Cloud Security Community site, googlecloudcommunity.com.

You can follow us on X, x.com/cloudsecpodcast. Tweet at us, email us, argue with us, and if we like or hate what we hear, we can invite you to the next episode.

See you on the next episode of the Cloud Security Podcast by Google.